On Sun, Jan 03, 2010 at 10:13:35PM -0800, Ryan Allen wrote:
> How does a modern linux kernel running software RAID do with CPU usage?

On any vaguely modern CPU, even RAID6 will probably be unnoticable.

There is one failure mode particular to software RAID that you should be
aware of. It may or may not be important to you.

Because the information regarding which disks are active in an array are
stored *on the disks* in software RAID, the potential exists for the boot
drive to wedge and be marked offline. If this situation is not detected
before the next boot, it may boot off that drive (which now contains stale
data) and your software may possibly do unexpected things with that old
data. On Windows, for example, it might put your entire Active Directory
domain in USN Rollback mode leaving everyone unable to login, or if you have
a database and some associated scheduled script that sends out emails based
on some date condition, you may find yourself apologizing for those emails.

Those are just two hypotheticals that I NEVER WANT TO HAVE HAPPEN TO ME
AGAIN. :)

Hardware RAID does not have this problem because the state of each disk is
stored in controller NVRAM instead.

Hardware RAID *does* have the additional problem that arrays will generally
only be readable with the same brand and family of controller. Not fun if
an old controller dies and you have to procure a used spare from
who-knows-where.
-- 
Robert Woodcock - [email protected]
"The only difference between me and a madman is that I am not mad."
        -- Salvador Dali

Reply via email to